RuntimeError: CUDA error: device-side assert triggered

[pacho69]

When ever i try to run with more then two classes cuda error occurred do i need to change some thing in notebook to run for more classes

[trbromley]

Hi @pacho69! It's hard to say without looking more closely at the notebook, but possible reasons for errors could be:

• A bug introduced while changing the code from two to more than two classes. Are you able to run with more than two classes on CPU (i.e., not trying to run on GPU)?
• Running with incompatible versions of, e.g., PyTorch. You can see the project requirements here, and can install them using pip install -r requirements.txt.
• Potentially some incompatibility with the hardware, e.g., need to update drivers. Can you train on GPU with two classes?

Could you share which notebook you are running?

[pacho69]

I have not tried to run more than two classes on CPU secondly i can train two classes on GPU

[trbromley]

Thanks! Looks like it's probably not a hardware issue then, maybe something in the code or getting the right versions of packages.

[pacho69]

As you can see all my requirements already met but the issue still persist

[trbromley]

Thanks! How about seeing if you can train on more than two classes with CPU. If it doesn't work, you'll know the problem is unrelated to GPUs and something to do with the code.